This is an automated email from the ASF dual-hosted git repository.
davidzollo pushed a commit to branch dev
in repository https://gitbox.apache.org/repos/asf/seatunnel.git
The following commit(s) were added to refs/heads/dev by this push:
new a9b95577f8 [Docs] Add Zeta job submission flow diagram (#11628)
a9b95577f8 is described below
commit a9b95577f82d25690c03d8db618236af4c5d80bd
Author: Daniel <[email protected]>
AuthorDate: Sun Aug 9 21:36:45 2026 +0800
[Docs] Add Zeta job submission flow diagram (#11628)
Co-authored-by: danielnadean
<[email protected]>
---
docs/en/architecture/engine/engine-architecture.md | 14 +++++++++++++-
...job_submission_classloader_task_execution_flow.png | Bin 0 -> 3684608 bytes
docs/zh/architecture/engine/engine-architecture.md | 14 +++++++++++++-
3 files changed, 26 insertions(+), 2 deletions(-)
diff --git a/docs/en/architecture/engine/engine-architecture.md
b/docs/en/architecture/engine/engine-architecture.md
index d0c54b3822..e324f8f8e2 100644
--- a/docs/en/architecture/engine/engine-architecture.md
+++ b/docs/en/architecture/engine/engine-architecture.md
@@ -76,7 +76,19 @@ flowchart TD
linkStyle default stroke:#5db8e2,stroke-width:2px;
```
-### 2.2 Core Components
+### 2.2 Job Submission, ClassLoader, and Task Dispatch Flow
+
+
+
+This flow fits the engine architecture section because it connects client
assembly, coordinator-owned scheduling, worker-side execution, and status
reporting in one view:
+
+1. The client resolves plugin jars locally and submits
`JobImmutableInformation`, which carries the logical DAG together with plugin
jar URLs and connector jar identifiers.
+2. `CoordinatorService` accepts the submission, records the pending job, and
returns the submit acknowledgement after `JobMaster.init()` and queueing are
complete.
+3. The coordinator-owned `Scheduler` polls `PendingJobQueue`, requests
resources through `preApplyResources()`, and triggers `JobMaster.run()` when
the `ResourceFuture` is ready.
+4. `JobMaster` expands the logical job into the pipeline-oriented
`ExecutionPlan` and `PhysicalPlan`, builds `TaskGroupImmutableInformation`, and
deploys it to workers through `PhysicalVertex.deploy()` and
`DeployTaskOperation`.
+5. Each worker resolves missing jars, creates child-first classloaders per
task, deserializes and initializes the task group inside
`TaskExecutionService`, executes the tasks, and reports deploy plus terminal
status back to `JobMaster` and `CoordinatorService`.
+
+### 2.3 Core Components
#### CoordinatorService
diff --git
a/docs/images/zeta_job_submission_classloader_task_execution_flow.png
b/docs/images/zeta_job_submission_classloader_task_execution_flow.png
new file mode 100644
index 0000000000..f6c9809b53
Binary files /dev/null and
b/docs/images/zeta_job_submission_classloader_task_execution_flow.png differ
diff --git a/docs/zh/architecture/engine/engine-architecture.md
b/docs/zh/architecture/engine/engine-architecture.md
index 383434f5ca..f0863984fa 100644
--- a/docs/zh/architecture/engine/engine-architecture.md
+++ b/docs/zh/architecture/engine/engine-architecture.md
@@ -78,7 +78,19 @@ flowchart TB
style worker fill:#081425,stroke:#8d7cf6,stroke-width:1.5px,color:#f8fbff;
```
-### 2.2 核心组件
+### 2.2 作业提交、类加载与任务下发流程
+
+
+
+这张图更适合放在引擎架构章节，因为它把客户端装配、Coordinator 侧调度、Worker 执行与状态回报串成了一条完整链路：
+
+1. 客户端先在本地解析插件目录，然后提交 `JobImmutableInformation`，其中同时包含逻辑 DAG、插件 jar URL 和连接器
jar 标识。
+2. `CoordinatorService` 接收提交请求后，会记录待调度作业，并在 `JobMaster.init()` 完成且作业入队后返回提交确认。
+3. Coordinator 持有的 `Scheduler` 会轮询 `PendingJobQueue`，通过 `preApplyResources()`
申请资源，并在 `ResourceFuture` 就绪后触发 `JobMaster.run()`。
+4. `JobMaster` 把逻辑作业展开为面向流水线的 `ExecutionPlan` 和 `PhysicalPlan`，构造
`TaskGroupImmutableInformation`，再通过 `PhysicalVertex.deploy()` 和
`DeployTaskOperation` 把任务下发到 Worker。
+5. 每个 Worker 会补齐缺失 jar、为每个任务创建 child-first 类加载器，在 `TaskExecutionService`
中反序列化并初始化任务组，执行任务，并把部署状态与最终状态回报给 `JobMaster` 和 `CoordinatorService`。
+
+### 2.3 核心组件
#### CoordinatorService
